|Publication number||US6988186 B2|
|Application number||US 09/894,260|
|Publication date||Jan 17, 2006|
|Filing date||Jun 28, 2001|
|Priority date||Jun 28, 2001|
|Also published as||US20030005263|
|Publication number||09894260, 894260, US 6988186 B2, US 6988186B2, US-B2-6988186, US6988186 B2, US6988186B2|
|Inventors||Richard James Eickemeyer, Steven R. Kunkel, Hung Q Le|
|Original Assignee||International Business Machines Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (11), Non-Patent Citations (11), Referenced by (49), Classifications (13), Legal Events (4)|
|External Links: USPTO, USPTO Assignment, Espacenet|
The present invention relates in general to an improved data processor architecture and in particular to a queue within the processor core to support simultaneous hardware multithreading.
From the standpoint of the computer's hardware, most systems operate in fundamentally the same manner. Computer processors actually perform very simple operations quickly, such as arithmetic, logical comparisons, and movement of data from one location to another. What is perceived by the user as a new or improved capability of a computer system, however, may actually be the machine performing the same simple operations at very high speeds. Continuing improvements to computer systems require that these processor systems be made ever faster.
One measurement of the overall speed of a computer system, also called the throughput, is measured as the number of operations performed per unit of time. Conceptually, the simplest of all possible improvements to system speed is to increase the clock speeds of the various components, particularly the clock speed of the processor. So that if everything runs twice as fast but otherwise works in exactly the same manner, the system will perform a given task in half the time. Computer processors which were constructed from discrete components years ago performed significantly faster reducing the size and number of components; eventually the entire processor was packaged as an integrated circuit on a single chip. The reduced size made it possible to increase the clock speed of the processor, and accordingly increase system speed.
Despite the enormous improvement in speed obtained from integrated circuitry, the demand for ever faster computer systems still exists. Hardware designers have been able to obtain still further improvements in speed by greater integration, by further reducing the size of the circuits, and by other techniques. Designers, however, think that physical size reductions cannot continue indefinitely and there are limits to continually increasing processor clock speeds. Attention has therefore been directed to other approaches for further improvements in overall throughput of the computer system.
Without changing the clock speed, it is still possible to improve system speed by using multiple processors. The modest cost of individual processors packaged on integrated circuit chips has made this practical. The use of slave processors considerably improves system speed by off-loading work from the central processing unit (CPU) to the slave processor. For instance, slave processors routinely execute repetitive and single special purpose programs, such as input/output device communications and control. It is also possible for multiple CPUs to be placed in a single computer system, typically a host-based system which serves multiple users simultaneously. Each of the different CPUs can separately execute a different task on behalf of a different user, thus increasing the overall speed of the system to execute multiple tasks simultaneously.
Coordinating the execution and delivery of results of various functions among multiple CPUs is a tricky business; not so much for slave I/O processors because their functions are pre-defined and limited but it is much more difficult to coordinate functions for multiple CPUs executing general purpose application programs. System designers often do not know the details of the programs in advance. Most application programs follow a single path or flow of steps performed by the processor. While it is sometimes possible to break up this single path into multiple parallel paths, a universal application for doing so is still being researched. Generally, breaking a lengthy task into smaller tasks for parallel processing by multiple processors is done by a software engineer writing code on a case-by-case basis. This ad hoc approach is especially problematic for executing commercial transactions which are not necessarily repetitive or predictable.
Thus, while multiple processors improve overall system performance, it is much more difficult to improve the speed at which a single task, such as an application program, executes. If the CPU clock speed is given, it is possible to further increase the speed of the CPU, i.e., the number of operations executed per second, by increasing the average number of operations executed per clock cycle. A common architecture for high performance, single-chip microprocessors is the reduced instruction set computer (RISC) architecture characterized by a small simplified set of frequently used instructions for rapid execution, those simple operations performed quickly as mentioned earlier. As semiconductor technology has advanced, the goal of RISC architecture has been to develop processors capable of executing one or more instructions on each clock cycle of the machine. Another approach to increase the average number of operations executed per clock cycle is to modify the hardware within the CPU. This throughput measure, clock cycles per instruction, is commonly used to characterize architectures for high performance processors.
Processor architectural concepts pioneered in high performance vector processors and mainframe computers of the 1970s, such as the CDC-6600 and Cray-1, are appearing in RISC microprocessors. Early RISC machines were very simple single-chip processors. As Very Large Scale Integrated (VLSI) technology improves, additional space becomes available on a semiconductor chip. Rather than increase the complexity of a processor architecture, most designers have decided to use the additional space to implement techniques to improve the execution of a single CPU. Two principal techniques utilized are on-chip caches and instruction pipelines. Cache memories store data that is frequently used near the processor and allow instruction execution to continue, in most cases, without waiting the full access time of a main memory. Some improvement has also been demonstrated with multiple execution units with hardware that speculatively looks ahead to find instructions to execute in parallel. Pipeline instruction execution allows subsequent instructions to begin execution before previously issued instructions have finished.
The superscalar processor is an example of a pipeline processor. The performance of a conventional RISC processor can be further increased in the superscalar computer and the Very Long Instruction Word (VLIW) computer, both of which execute more than one instruction in parallel per processor cycle. In these architectures, multiple functional or execution units are connected in parallel to run multiple pipelines. The name implies that these processors are scalar processors capable of executing more than one instruction in each cycle. The elements of superscalar pipelined execution may include an instruction fetch unit to fetch more than one instruction at a time from a cache memory, instruction decoding logic to determine if instructions are independent and can be executed simultaneously, and sufficient execution units to execute several instructions at one time. The execution units may also be pipelined, e.g., floating point adders or multipliers may have a cycle time for each execution stage that matches the cycle times for the fetch and decode stages.
In a superscalar architecture, instructions may be completed in-order and/or out-of-order. In-order completion means no instruction can complete before all instructions dispatched ahead of it have been completed. Out-of-order completion means that an instruction is allowed to complete before all instructions ahead of it have been completed, as long as a predefined rules are satisfied. Within a pipelined superscalar processor, instructions are first fetched, decoded and then buffered. Instructions can be dispatched to execution units as resources and operands become available. Additionally, instructions can be fetched and dispatched speculatively based on predictions about branches taken. The result is a pool of instructions in varying stages of execution, none of which have completed by writing final results. These instructions in different stages of interim execution may be stored in a variety of queues used to maintain the in-order appearance of execution. As resources become available and branches are resolved, the instructions are retrieved from their respective queue and “retired” in program order thus preserving the appearance of a machine that executes the instructions in program order.
Another technique called hardware multithreading independently executes smaller sequences of instructions called threads or contexts in a single processor. When a CPU, for any of a number of reasons, stalls and cannot continue processing or executing one of these threads, it switches to and executes another thread. The term “multithreading” as defined in the computer architecture community is not the same as the software use of the term in which one task is subdivided into multiple related threads. Software multithreading substantially involves the operating system which manipulates and saves data from registers to main memory and maintains the program order of related and dependent instructions before switching tasks. Software multithreading does not require nor is it concerned with hardware multithreading and vice versa. Hardware multithreading manipulates hardware architected registers and execution units and pipelined processors within the processor core to maintain the state of one or more independently executing sets of instructions, called threads, in the processor hardware. Threads could be derived from, for example, different tasks in a multitasking system, different threads compiled from a software multithreading system, or from different I/O processors. What makes hardware multithreading unique and different from all these systems, however, is that more than one thread is independently maintained in a processor's registers.
Hardware multithreading takes on a myriad of forms. Multithreading permits processors having either non-pipelined or pipelined architectures to do useful work on more than one thread in the processor's registers. One form of multithreading, sometimes referred to as coarse-grained multithreading, is to execute one thread until the executing thread experiences a long latency event, such as retrieving data and/or instructions from memory or a processor interrupt, etc. Fine-grained multithreading, on the other hand, interleaves or switches threads on a cycle-by-cycle basis. Simultaneous hardware multithreading maintains N threads, or N states, in parallel in the processor and simultaneously executes N threads in parallel. Replicating processor registers for each of N threads results in some of the following registers being replicated N times: general purpose registers, floating point registers, condition registers, floating point status and control registers, count registers, link registers, exception registers, save/restore registers, special purpose registers, etc. Special buffers, such as a segment lookaside buffer, may be replicated but if not, each entry can be tagged with the thread number and flushed on every thread switch. Also, some branch prediction mechanisms, e.g., the correlation register and the return stack, may also be replicated.
Multithreading may also take on features of one or all of the forms, picking and choosing particular features for particular attributes. Not all of the processor's features need be replicated for each thread and there may be some shared and some replicated registers. Stages in the pipeline may either be separate or shared. Preferably, there may be no need to replicate some of the larger functions of the processor such as the level one instruction cache, level one data cache, instruction buffer, store queue, instruction dispatcher, functional or execution units, pipelines, translation lookaside buffer (TLB), and branch history table.
With respect to the threads there may be private or separate resources, or shared resources. Private resources simplify the management in that pointers from different threads address different data structures and may permit advantageous placement of the queues in different parts of the chip. Some private resources are registers or queues dedicated for the exclusive use of a particular thread. Another example of a private resource may be a split queue or registers having reserved spaces for each thread. An example of a partitioned queue is set forth in U.S. patent application Ser. No. 09/645,081 filed 24 Aug. 2000 entitled Method for Implementing a Variable-Partitioned Queue for Simultaneous Multithreaded Processors, which application is owned by the assignee herein and which is hereby incorporated by reference in its entirety.
The combination of out-of-order processing with simultaneous multithreading reveal some of the classical dilemmas of processor architecture and design. In order to proceed quickly, each thread may maintain its own out-of order queues of instructions in varying stages of completion, i.e., have its own private resources. Yet, the additional resources required for each thread are expensive and may be cumbersome in terms of space and power. Separate resources, moreover, lack the flexibility required for dynamic simultaneous multithreading. Execution of a thread of high priority, for example, may demand significantly more hardware registers and queues and other resources than has been architected for a single thread. Conversely, a different thread may not utilize all the space it has been allotted and so there are wasted resources. The partitioned queues as above lack the spontaneous flexibility and responsiveness that may be required by simultaneous multithreading.
In accordance with a preferred embodiment of the invention, the use of shared resources for the various queue yields performance benefits similar to the split queue structure without increasing the chip area over the split queue case. But, normally shared resources present their own set of problems. If no dispatch flush can be performed on a shared resource, processing is stalled because the shared resource, such as the register renamed pool, is full or otherwise blocked. Under these circumstances all threads are blocked such that no processing of any thread can occur and either a normal flush would occur or the processor would wait until the stalling condition is resolved. To complicate matters even further, there are certain conditions under which a dispatch flush cannot occur, such as when the stalled instruction is of a group of instructions generated from a decoded multiple or more complex instruction but the stalled instruction is not the first of the group.
There is thus a need to accommodate dynamic out-of-order processing of multiple threads in a processor architecture.
These needs and others that will become apparent to one skilled in the art are satisfied by a resource queue, comprising: a plurality of entries, each entry having unique resources required for information processing in which the plurality of entries is allocated amongst a plurality of independent hardware threads such that the resources of more than one thread may be within the queue and the entries allocated to one thread may be interspersed among the entries allocated to another thread.
The first entry of one thread may wrap around the last entry of the same thread. Each thread may have a head pointer and a tail pointer wherein the head pointer is the first entry of the thread and the tail pointer is the last entry of the thread, and one of the unique resources is a bank number to indicate how many times the head pointer has wrapped around the tail pointer in order to maintain an order of the resources for the thread. There may also be a free pointer for a thread indicating an entry in the queue available for resources of the at least one thread.
The information processing may occur within an out-of-order computer processor, and the resource queue may further comprise a load reorder queue and/or a store reorder queue and/or a global completion table and or a branch information queue.
The invention may also be considered a resource queue in an out-of-order multithreaded computer processor, comprising: a load reorder queue; a store reorder queue; a global completion table; a branch information queue, wherein at least one of the queues comprises: a plurality of entries, each entry having unique resources required for information processing; the plurality of entries allocated amongst a plurality of independent hardware threads such that the resources of more than one thread may be within the queue; and the entries allocated to one thread may be interspersed among the entries allocated to another thread; and a first entry of one thread being capable of wrapping around the last entry of the same thread; and at least one thread having a head pointer and a tail pointer wherein the head pointer is the first entry of the thread and the tail pointer is the last entry of the thread, and a bank number indicates how many times the head pointer has wrapped around the tail pointer; and at least one free pointer for the at least one thread indicating an entry in the queue is available for resources of the thread.
The invention is also a method of allocating a shared resource queue for multithreaded electronic data processing, comprising the steps of determining if the shared resource queue is empty for a particular thread; finding the first entry of a particular thread; determining if the first entry and a free entry of the particular thread are the same; if, not advancing the first entry to the free entry; incrementing a bank number if the first entry passes the last entry before it finds the free entry; and allocating the next free entry by storing resources for the particular thread.
The method may further comprise deallocating multithreaded resources in the shared resource queue, comprising the steps of locating the last entry in the shared resource queue pertaining to the particular thread; determining if the last entry is also the first entry for the particular thread; if not, finding the next entry pertaining to the particular thread; determining if the bank number of the next entry is the same as the last entry and if so, deallocating the next entry by marking the resources as invalid; and if not, then skipping over the next entry and decrementing the bank number; and finding the next previous entry pertaining to the particular thread.
The shared resource queue may also be flushed by setting a flush point indicative of an oldest entry to be deallocated pertaining to the particular thread, and invalidating all entries between the head pointer and the flush point which have the same and greater bank number than the bank number of the flush point.
The invention is also considered a shared resource mechanism in a hardware multithreaded pipeline processor, the pipeline processor simultaneously processing a plurality of threads, the shared resource mechanism comprising: a dispatch stage of the pipeline processor; at least one shared resource queue connected to the dispatch stage; dispatch control logic connected to the dispatch stage and to at least one shared resource queue; and an issue queue of the pipeline processor connected to the dispatch stage and to the at least one shared resource queue, wherein the at least one shared resource queue allocates and deallocates resources for at least two threads passing into the issue queues in response to the dispatch control logic.
The invention is also an apparatus to enhance processor efficiency, comprising: means to fetch instructions from a plurality of threads into a hardware multithreaded pipeline processor; means to distinguish the instructions into one of a plurality of threads; means to decode the instructions; means to allocate a plurality of entries in at least one shared resource between at least two of the plurality of threads; means to determine if the instructions have sufficient private resources and at least one shared resource queue for dispatching the instructions; means to dispatch the instructions; means to deallocate the entries in the shared resource when one of the threads are dispatched; means to execute the instructions and the resources for the one of the threads. The apparatus may further comprises a means to flush the shared resource of all of the entries pertaining to a particular thread.
The invention is also a computer processing system, comprising: a central processing unit; a semiconductor memory unit attached to the central processing unit; at least one memory drive capable of having removable memory; a keyboard/pointing device controller attached to the central processing unit for attachment to a keyboard and/or a pointing device for a user to interact with the computer processing system; a plurality of adapters connected to the central processing unit to connect to at least one input/output device for purposes of communicating with other computers, networks, peripheral devices, and display devices; a hardware multithreading pipelined processor within the central processing unit to process at least two independent threads of execution, the pipelined processor comprising a fetch stage, a decode stage, and a dispatch stage; and at least one shared resource queue within the central processing unit, the shared resource queue having a plurality of entries pertaining to more than one thread in which entries pertaining to different threads are interspersed among each other. In addition, a first entry of one thread may be located after a last entry of the one thread. Further, the hardware multithreaded pipelined processor in the central processing unit may be an out-of-order processor.
The invention is best understood with reference the Drawing and the detailed description of the invention which follows.
Referring now to the Drawing wherein like numerals refer to the same or similar elements throughout and in particular with reference to
CPU 126 is a general-purpose programmable multithreaded processor, executing instructions stored in memory 158. While a single CPU having multithreaded capabilities is shown in
Memory 158 is a random-access semiconductor memory (RAM) for storing data and programs; memory is shown conceptually as a single monolithic entity, it being understood that memory is often arranged in a hierarchy of caches and other memory devices. RAM 158 typically comprises a number of individual volatile memory modules that store segments of operating system and application software while power is supplied to processing unit 112. The software segments may be partitioned into one or more virtual memory pages that each contain a uniform number of virtual memory addresses. When the execution of software requires more pages of virtual memory than can be stored within RAM 158, pages that are not currently needed are swapped with the required pages, which are stored within non-volatile storage devices 122, 123, or 124. Data storage 123 and 124 preferably comprise one or more rotating magnetic or optical hard disk drive units, although other types of data storage could be used.
Keyboard/pointing-device controller 184 interfaces processing unit 112 with a keyboard and graphical pointing device. In an alternative embodiment, there may be a separate controller for the keyboard and the graphical pointing device and/or other input devices may be supported, such as microphones, voice response units, etc. Display device adapter 170 translates data from CPU 126 into video, audio, or other signals utilized to drive a display or other output device. Device adapter 170 may support the attachment of a single or multiple terminals, and may be implemented as one or multiple electronic circuit cards or other units.
Processing unit 112 may include network-adapter 185, media communications interface 187, and parallel-port adapter 176, all of which facilitate communication between processing unit 112 and peripheral devices or other data processing system. Parallel port adapter 176 may transmit printer-control signals to a printer through a parallel port. Network-adapter 185 may connect processing unit 112 to a local area network (LAN). A LAN provides a user of processing unit 112 with a means of electronically communicating information, including software, with a remote computer or a network logical storage device. In addition, a LAN supports distributed processing which enables processing unit 112 to share a task with other data processing systems linked to the LAN. For example, processing unit 112 may be connected to a local server computer system via a LAN using an Ethernet, Token Ring, or other protocol, the server in turn being connected to the Internet. Media communications interface 187 may comprise a modem connected to a telephone line or other higher bandwidth interfaces through which an Internet access provider or on-line service provider is reached. Media communications interface 187 may interface with cable television, wireless communications, or high bandwidth communications lines and other types of connection. An on-line service may provide software that can be downloaded into processing unit 112 via media communications interface 187. Furthermore, through the media communications interface 187, processing unit 112 can access other sources of software such as a server, electronic mail, or an electronic bulletin board, and the Internet or world wide web.
The processor as disclosed in the above incorporated applications may be modified as described below to enable simultaneous out-of-order hardware multithreading operations. Multithreading operations in the context of the invention means hardware multithreading in which sequences of instructions, i.e., threads, execute independently from other threads, and in which hardware architected registers, execution units, and pipelined processors maintain the state of one or more independently executing sets of instructions, called threads, in the processor core hardware. With simultaneous multithreading, two or more threads are simultaneously active in the processor's pipeline. Thus, the processors' pipeline(s) are able to perform useful work on different threads when a processor pipeline stall condition is detected for one thread. Furthermore, while one embodiment of the invention will be described in the context of dual multithreading operations in which only two threads are in the processor's pipeline, registers, and queues at any one time, the principles and context of the shared queue of the invention pertain to having more than two threads in different stages of the processor's pipeline and the architected registers and queues. One of skill in the art will appreciate, moreover, that multithreaded pipelined processor architectures not having out-of-order execution or the particular sets of registers and queues as described in the incorporated patent applications can also take advantage of the shared queue feature of the present invention.
The block diagram of a multithreaded pipeline processor of
The decode unit 218 may require multiple cycles to complete its function and accordingly, may have multiple pipelines 218 a, 218 b, etc. Preferably each stage 218 a and stage 218 b has distinct threads. In the decode unit 218, complex instructions may be simplified or represented in a different form for easier processing by subsequent processor pipeline stages. Other events that may occur in the decode unit 218 include the reshuffling or expansion of bits in instruction fields, extraction of information from various fields for, e.g., branch prediction or creating groups of instructions. Some instructions, such as load multiple or store multiple instructions, are very complex and are processed by breaking the instruction into a series of simpler operations or instructions, called microcode, during decode.
From the decode unit 218, instructions are forwarded to the dispatch unit 220. The dispatch unit 220 may receive control signals from the dispatch control 240 in accordance with the referenced applications. At the dispatch unit 220 of the processor pipeline, all resources, queues, and renamed pools are checked to determine if they are available for the instructions within the dispatch unit 220. Different instructions have different requirements and all of those requirements must be met before the instruction is dispatched beyond the dispatch unit 220. The dispatch control 240 and the dispatch unit 220 controls the dispatch of microcoded or other complex instructions that have been decoded into a multitude of simpler instructions, as described above. The processor pipeline, in one embodiment, typically will not dispatch in the middle of a microcoded instruction group; the first instruction of the microcode must be dispatched successfully and the subsequent instructions are dispatched in order. A multithread embodiment of the invention may presume that during any given processor clock cycle an instruction from only one thread is dispatched from the dispatch unit 220 to the issue queues 222.
From the dispatch unit 220, instructions enter the issue queues 222. There may be multiple issue queues 222 a and 222 b and more than one thread may be present in each issue queue 222 a, 222 b in accordance with an embodiment of the shared queues as described herein, or there may be one issue queue per thread depending upon choice of architecture. The issue queues 222 may receive control signals from the completion control logic 236, from the dispatch control 240, and from a combination of various queues which may include, but which are not limited to, a non-renamed register tracking mechanism 242, a load reorder queue (LRQ) 244, a store reorder queue (SRQ) 246, a global completion table (GCT) 248, and a rename pools 250. The LRQ 244, the SRQ 246, and/or the GCT 248 may be split between threads, may be shared amongst the threads in accordance with an embodiment of the shared queue as described herein; or separate queues may exist for each thread as will be discussed in the context of separate or shared resources below. For tracking purposes, instructions may be tracked singly or in groups in the GCT 248 to maintain the order of instructions. The LRQ 244 and the SRQ 246 may maintain the order of the load and store instructions, respectively, as well as maintaining addresses for the program order. The non-renamed register tracking mechanism 242 may track instructions in such registers as special purpose registers, etc. The instructions are dispatched on yet another machine cycle to the designated execution unit which may be one or more condition register units 224, branch units 226, fixed point units 228, floating point units 230, or load/store units 232 which load and store data from and to the data cache 234.
The successful completion of execution of an instruction or, on the other hand, mispredicted branches or notification of errors which may have occurred in the execution units are forwarded to the completion control logic 236, which may generate and transmit a refetch signal to any of a plurality of queues, the non-renamed register tracking mechanism 242, the LRQ 244, the SRQ 246, the GCT 248, or the renamed pools 250. Alternatively, recovery and/or flush techniques may occur in response to the output of the completion control logic 236.
As discussed, though, separate data structures and queues are not as efficient as a structure shared amongst the threads because either the structure is fixed or changes so slowly to be unable to accommodate dynamic and responsive processing. Thus, the preferred embodiment of the invention further contemplates that the data structures and registers be architected as shared resources. Shared resources are those processor registers and queues which can be shared by either thread either separately or at the same time. In some circumstances, the non-renamed tracking register 242, the LRQ 244, the SRQ 246, the GCT 248, the register renamed pools 250, the issue queues 222, the branch information queue (BIQ) 252 may be shared amongst threads. Alternatively, some resources may be shared while others may be private, e.g., the GCT 248 may be shared while the LRQ 244 and the SRQ 246 may be private depending upon the particular architecture. The use of shared resources for the various queues yields performance benefits without increasing the chip area over the shared queue case.
The queue elements or entries 310–332 are organized into a first-in first-out (FIFO) queue 300. There is a head pointer and a tail pointer for each thread. The head and tail pointers of one thread are independent from those of another thread. An entry can be occupied by only one thread at a particular time. To improve queue allocation time, each thread also has a free pointer which points to the next entry that the thread could use for allocation. Referring specifically to
In a typical queue the valid entries for one thread are located between the head and tail pointers for that thread when viewed as a circular queue. There are entries between the head and tail pointers of a particular thread but that are outside the range of another thread's head and tail pointer and hence unusable by the other thread. There may be many unusable entries if there are multiple threads. In a preferred embodiment of the invention, as shown in
In order to best understand and take advantage of this wrap-around features, it is useful to establish some basic premises: the tail pointer indicates the oldest entry. As the tail pointer advances, entries are deallocated, i.e., made invalid. The head pointer indicates the newest entry; as the head advances, entries are allocated, i.e., made valid. The free pointer points to the next entry to be allocated, i.e., the next entry for the head. If there is no entry that can be allocated, the free pointer is the same as the head pointer. As the head and tail pointers move in reaction to allocation and deallocation of entries, the head pointer can pass the tail pointer and the tail pointer can pass the head pointer, in either direction, thus preventing unusable entries in the queue.
Recall that the process in
If the current entry, however, is not the same as the head pointer of a particular thread, in step 620, then the adjacent entry is selected as in step 622. In the queue structure of
If, however, the current entry is not the same as the tail pointer, then in step 722 the adjacent entry in an opposing direction to that used in
Starting at step 810 of
Starting at step 820, the counter “x” for the entry is set to zero and a free pointer is set as the head pointer, in step 824. On the other hand, if at step 810, the queue already contains entries pertaining to the thread under consideration, then in step 822, the counter “x” is set to the bank number of the entry at the head and again, the free pointer in that entry having the head pointer for the thread is set, as in step 824. In step 826 the process advances the free pointer to the adjacent entry and establishes whether that entry is the tail pointer, as in step 828. If so, at step 830, the process checks if the head entry is at the maximum bank number. If so, then the free pointer is set to be a head pointer in step 832 and the queue does not have enough banks for the particular thread under consideration.
If, in step 828, the entry at the free pointer is not at the tail pointer then it proceeds to check if the entry is the head pointer in step 834. If so, then the queue is full and process is completed as in step 890. If the entry at the free pointer is not at the head pointer, then, in step 836, the entry is checked to see if the entry at the free pointer is valid. If so, that entry is already occupied. If the entry having the free pointer is not valid, it means that the entry is available for the thread under consideration and the process completes in step 890. If the entry is valid, the free pointer moves to the adjacent entry at step 826.
In more detail now, the process begins at step 910 and inquires if the queue contains any entries pertaining to the thread under consideration (step 902). If so, in step 912 the counters “y” and “z” are set to “0”, and the counter “x” is set to bank of the head pointer. If there are no preexisting entries in the queue as determined in step 902, then in step 914, the queue is initialized for that thread by setting the counters “x” and “z” to “0”, and counter “y” to “1”. In any event, the process inquires at step 916 if the head pointer and the free pointer for a particular thread are at the same entry. If so, this indicates that the queue is full and the process terminates to step 990. If, however, the queue is not full, then at step 918, the process advances the head pointer to the adjacent entry and asks if that entry is also the tail pointer as in step 920. If so, then in step 922, it means that the head pointer has passed the tail pointer and so in step 922, the counter “z” is set to one and proceeds to step 924. If the head pointer is not the same as the tail pointer (step 920), the process inquires at step 924 if the entry is at the free pointer. If not, then the process loops back to step 918 to move the head pointer to the adjacent entry. If, however, the entry is at the free pointer in step 924, then in step 926 the process sets the bank of the new entry to either the bank of the original head entry or one higher if the tail pointer was passed.
In step 928, the process queries the value of “y” to see if the queue was originally empty at start. If so, the tail pointer is set to the head pointer indicating that the queue has only one entry pertaining to a particular thread as in step 930. Now the new entry is initialized and the entry is set to contain valid information, pertain to the particular thread, and store the information in the entry for successful allocation, as in step 932. The process then terminates at step 990.
At step 1010, the process begins and the entry at the tail pointer is made invalid in step 1012. The process then inquires if the entry having the tail pointer also contains the head pointer in step 1014. If so, it means that there was only one entry in the queue pertaining to the thread under consideration and the process completes in step 1090 because all the entries have been deallocated by marking the valid field to zero. If, however, there are more entries pertaining to the particular thread, then in step 1016, the entry counter “y” is set to hold the tail pointer. The process then finds the next entry from that tail pointer pertaining to that thread, as in
In step 1110, the process starts and in step 1102 determines if the queue has any entries for the particular thread. If the thread is empty for the thread, the process terminates at step 1190. If, however, the queue has entries for the thread under consideration, then at step 1112, the counter “flush” is set to the oldest entry to be flushed. In step 1114, the process determines if the flushed entry is at the tail pointer. If so, the queue for the thread will be empty, so the head pointer is set to the tail pointer. If any entries are not flushed, as determined by step 1114, then in step 1116, the process finds the previous entry as in
If, in step 1126, the bank counter is not equal to the bank that is being flushed, the process checks to see if the bank counter is greater than the bank of the entries being flushed, as in step 1136. If the bank counter is greater than the bank counter being flushed, then the entry is flushed in step 1138 by setting the entry to be invalid and the bank counter to be zero.
If, in step 1130, the entry is the same as the counter “flush” then the counter “z” is set to one to indicate that we found the flush entry in step 1132. The entry is made invalid in step 1134 and the process advances to step 1140.
Note that there are many repetitive steps throughout the figures that easily lend themselves to modular software loops or preferably, if manifested in hardware, to route the digital signals through the same logic.
Thus, while the invention has been described with respect to preferred and alternate embodiments, it is to be understood that the invention is not limited to processors which have only out-of-order processing but is particularly useful in such applications. The invention is intended to be manifested in the following claims.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5465120 *||Feb 7, 1994||Nov 7, 1995||The Grass Valley Group, Inc.||Spiral buffer for non-linear editing|
|US5623608 *||Nov 14, 1994||Apr 22, 1997||International Business Machines Corporation||Method and apparatus for adaptive circular predictive buffer management|
|US6075931 *||Jun 25, 1997||Jun 13, 2000||Sun Microsystems, Inc.||Method for efficient implementation of multi-ported logic FIFO structures in a processor|
|US6311261 *||Sep 15, 1997||Oct 30, 2001||Georgia Tech Research Corporation||Apparatus and method for improving superscalar processors|
|US6353829 *||Dec 23, 1998||Mar 5, 2002||Cray Inc.||Method and system for memory allocation in a multiprocessing environment|
|US6507921 *||Oct 1, 1999||Jan 14, 2003||Texas Instruments Incorporated||Trace fifo management|
|US6629271 *||Dec 28, 1999||Sep 30, 2003||Intel Corporation||Technique for synchronizing faults in a processor having a replay system|
|US20010032307 *||Dec 30, 1998||Oct 18, 2001||Joseph Rohlman||Micro-instruction queue for a microprocessor instruction pipeline|
|US20020078317 *||Dec 19, 2000||Jun 20, 2002||Matsushita Electric Industrial Co., Ltd.||First-in, first-out (FIFO) memory with moving boundary|
|US20030061258 *||Sep 18, 2002||Mar 27, 2003||Dion Rodgers||Method and apparatus for processing an event occurrence for at least one thread within a multithreaded processor|
|US20030105944 *||Nov 11, 2002||Jun 5, 2003||Hewlett-Packard Development Company||Method and apparatus to quiesce a portion of a simultaneous multithreaded central processing unit|
|1||*||Daniele Folegnani, Antonio González, "Energy-effective issue logic", Proceedings of the 28th annual international symposium on Computer architecture, p. 230-239, Jun. 30-Jul. 4, 2001, Göteborg, Sweden.|
|2||*||Gulati, M., Bagherzadeh, N. "Performance Study of a Multithreaded Superscaler Microprocessor". Proc. of the 2 nd International Symposium on High-Performance Computer Architectures, Feb. 1996, 291-301.|
|3||*||Ponomarev, D., Kucuk, G., Ghose, K., "Dynamic Allocation of Datapath Resources for Low Power", in Proc. of Workshop on Complexity-Effective Design, held in conjunction with ISCA-28, Jun. 2001.|
|4||*||S. Reinhardt and S. Mukherjee, "Transient fault detection via simultaneous multithreading" 27th Int'l Symp. on Computer Architecture, Jun. 2000.|
|5||U.S. Appl. No. 09/108,160 entitled System and Method for Dispatching Groups of Instructions, filed Jun. 30, 1998.|
|6||U.S. Appl. No. 09/213,323 entitled System and Method for Permitting Out-of-Order Execution of Load Instructions, filed Dec. 16, 1998.|
|7||U.S. Appl. No. 09/213,331 entitled System and Method for Permitting Out-of-Order Execution of Load and Store Instructions, filed Dec. 16, 1998.|
|8||U.S. Appl. No. 09/332,413 entitled Method and System for Restoring a Processor State Within a Data Processing System in which instructions are Tracked in Groups, filed Jul. 14, 1999.|
|9||U.S. Appl. No. 09/434,095 entitled System and Method for Managing the Execution of Instruction Groups Having Multiple Executable Instructions, filed Nov. 5, 1999.|
|10||U.S. Appl. No. 09/564,930 entitled Selective Flush of Shared and Other Pipelined Stages in a Multithreaded Processor, filed May 4, 2000.|
|11||U.S. Appl. No. 09/645,08 entitled Method for Implementing a Variable-Partitioned Queue for Simultaneous Multithreaded Processors, filed Aug. 24, 2000.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7194603 *||Apr 23, 2003||Mar 20, 2007||International Business Machines Corporation||SMT flush arbitration|
|US7266650 *||Nov 12, 2004||Sep 4, 2007||International Business Machines Corporation||Method, apparatus, and computer program product for implementing enhanced circular queue using loop counts|
|US7366884 *||Feb 25, 2002||Apr 29, 2008||Agere Systems Inc.||Context switching system for a multi-thread execution pipeline loop and method of operation thereof|
|US7458078||Nov 6, 2003||Nov 25, 2008||International Business Machines Corporation||Apparatus and method for autonomic hardware assisted thread stack tracking|
|US7802032||Nov 13, 2006||Sep 21, 2010||International Business Machines Corporation||Concurrent, non-blocking, lock-free queue and method, apparatus, and computer program product for implementing same|
|US7814243||Jun 1, 2007||Oct 12, 2010||Sonics, Inc.||Shared storage for multi-threaded ordered queues in an interconnect|
|US7987453 *||Mar 18, 2004||Jul 26, 2011||International Business Machines Corporation||Method and apparatus for determining computer program flows autonomically using hardware assisted thread stack tracking and cataloged symbolic data|
|US8006075 *||May 21, 2009||Aug 23, 2011||Oracle America, Inc.||Dynamically allocated store queue for a multithreaded processor|
|US8087034 *||Oct 26, 2009||Dec 27, 2011||Steven J. Frank||Virtual processor methods and apparatus with unified event notification and consumer-produced memory operations|
|US8099566||May 15, 2009||Jan 17, 2012||Oracle America, Inc.||Load/store ordering in a threaded out-of-order processor|
|US8166214||Jan 8, 2010||Apr 24, 2012||Sonics, Inc.||Shared storage for multi-threaded ordered queues in an interconnect|
|US8195739 *||Oct 31, 2007||Jun 5, 2012||Tibco Software Inc.||Adaptive polling|
|US8230177||May 28, 2009||Jul 24, 2012||Oracle America, Inc.||Store prefetching via store queue lookahead|
|US8271997||Feb 4, 2010||Sep 18, 2012||Steven J. Frank||General purpose embedded processor|
|US8335911 *||Sep 30, 2009||Dec 18, 2012||Oracle America, Inc.||Dynamic allocation of resources in a threaded, heterogeneous processor|
|US8381037||Oct 9, 2003||Feb 19, 2013||International Business Machines Corporation||Method and system for autonomic execution path selection in an application|
|US8601446 *||Oct 22, 2008||Dec 3, 2013||Wipro Limited||Complexity dependency change impact system and method|
|US8615619||Jan 11, 2011||Dec 24, 2013||International Business Machines Corporation||Qualifying collection of performance monitoring events by types of interrupt when interrupt occurs|
|US8621487||Nov 14, 2011||Dec 31, 2013||Steven J. Frank||Virtual processor methods and apparatus with unified event notification and consumer-producer memory operations|
|US8650384 *||Apr 27, 2010||Feb 11, 2014||Samsung Electronics Co., Ltd.||Method and system for dynamically parallelizing application program|
|US8689190||Jan 29, 2008||Apr 1, 2014||International Business Machines Corporation||Counting instruction execution and data accesses|
|US8694697 *||Apr 27, 2006||Apr 8, 2014||Nvidia Corporation||Rescindable instruction dispatcher|
|US8782664||Jan 11, 2012||Jul 15, 2014||International Business Machines Corporation||Autonomic hardware assist for patching code|
|US8938598 *||Jul 6, 2011||Jan 20, 2015||Nvidia Corporation||Facilitating simultaneous submission to a multi-producer queue by multiple threads with inner and outer pointers|
|US8972995||Aug 6, 2010||Mar 3, 2015||Sonics, Inc.||Apparatus and methods to concurrently perform per-thread as well as per-tag memory access scheduling within a thread and across two or more threads|
|US9189277||Jul 22, 2013||Nov 17, 2015||Samsung Electronics Co., Ltd.||Method and system for dynamically parallelizing application program|
|US20030163675 *||Feb 25, 2002||Aug 28, 2003||Agere Systems Guardian Corp.||Context switching system for a multi-thread execution pipeline loop and method of operation thereof|
|US20030214326 *||Feb 11, 2003||Nov 20, 2003||Craimer Stephen G.||Distributed dynamically optimizable processing communications and storage system|
|US20040215938 *||Apr 23, 2003||Oct 28, 2004||International Business Machines Corporation||SMT flush arbitration|
|US20050081107 *||Oct 9, 2003||Apr 14, 2005||International Business Machines Corporation||Method and system for autonomic execution path selection in an application|
|US20050102673 *||Nov 6, 2003||May 12, 2005||International Business Machines Corporation||Apparatus and method for autonomic hardware assisted thread stack tracking|
|US20050138333 *||Dec 19, 2003||Jun 23, 2005||Samra Nicholas G.||Thread switching mechanism|
|US20050149700 *||Dec 19, 2003||Jul 7, 2005||Samra Nicholas G.||Virtual multithreading translation mechanism including retrofit capability|
|US20050210454 *||Mar 18, 2004||Sep 22, 2005||International Business Machines Corporation||Method and apparatus for determining computer program flows autonomically using hardware assisted thread stack tracking and cataloged symbolic data|
|US20060106749 *||Nov 12, 2004||May 18, 2006||International Business Machines Corporation||Method, apparatus, and computer program product for implementing enhanced circular queue using loop counts|
|US20080112423 *||Nov 13, 2006||May 15, 2008||David Alan Christenson||Concurrent, Non-Blocking, Lock-Free Queue and Method, Apparatus, and Computer Program Product for Implementing Same|
|US20080301708 *||Jun 1, 2007||Dec 4, 2008||Hamilton Stephen W||Shared storage for multi-threaded ordered queues in an interconnect|
|US20080313345 *||Oct 31, 2007||Dec 18, 2008||Datasynapse, Inc.||Adaptive polling|
|US20100100874 *||Oct 22, 2008||Apr 22, 2010||Ajikumar Thaitharanikarthu Narayanan||Complexity dependency change impact system and method|
|US20100115196 *||Jan 8, 2010||May 6, 2010||Sonics, Inc.||Shared storage for multi-threaded ordered queues in an interconnect|
|US20100162028 *||Oct 26, 2009||Jun 24, 2010||Sharp Kabushiki Kaisha Corporation||Virtual processor methods and apparatus with unified event notification and consumer-produced memory operations|
|US20100281489 *||Apr 27, 2010||Nov 4, 2010||Samsung Electronics Co., Ltd.||Method and system for dynamically parallelizing application program|
|US20100293347 *||Nov 18, 2010||Luttrell Mark A||Load/store ordering in a threaded out-of-order processor|
|US20100299499 *||Sep 30, 2009||Nov 25, 2010||Golla Robert T||Dynamic allocation of resources in a threaded, heterogeneous processor|
|US20100299508 *||May 21, 2009||Nov 25, 2010||Luttrell Mark A||Dynamically allocated store queue for a multithreaded processor|
|US20120233441 *||Sep 13, 2012||Barreh Jama I||Multi-threaded instruction buffer design|
|US20130014118 *||Jan 10, 2013||Stephen Jones||Simultaneous submission to a multi-producer queue by multiple threads|
|EP2156305A1 *||May 30, 2008||Feb 24, 2010||Sonics, INC.||Shared storage for multi-threaded ordered queues in an interconnect|
|WO2008148132A1 *||May 30, 2008||Dec 4, 2008||Hamilton Stephen W||Shared storage for multi-threaded ordered queues in an interconnect|
|U.S. Classification||712/217, 712/E09.053, 712/228, 712/E09.055|
|International Classification||G06F9/00, G06F15/167, G06F9/38|
|Cooperative Classification||G06F9/3802, G06F9/3851, G06F9/3814|
|European Classification||G06F9/38B, G06F9/38B8, G06F9/38E4|
|Jun 28, 2001||AS||Assignment|
Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION, NEW Y
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:EICKEMEYER, RICHARD JAMES;KUNKEL, STEVEN R.;LE, HUNG QUI;REEL/FRAME:011970/0894;SIGNING DATES FROM 20010618 TO 20010622
|Jul 27, 2009||REMI||Maintenance fee reminder mailed|
|Jan 17, 2010||LAPS||Lapse for failure to pay maintenance fees|
|Mar 9, 2010||FP||Expired due to failure to pay maintenance fee|
Effective date: 20100117